Pollutin' in Memphis (to the tune of "Walking in Memphis")
[Verse 1]
Put on my Tesla flame thrower
And I tweeted my plan
Touched down in the land of the Delta Blues
With a methane-powered scam
W.C. Handy
Won't you wheeze down over me?
Yeah, I've got a first-class subsidy
But I'm as toxic as can be
[Chorus]
Then I'm pollutin' in Memphis
I was spewin' with my turbines ten feet off of Beale
Pollutin' in Memphis
But do y'all smell the way I feel?
[Verse 2]
Saw the ghost of Elvis
Gaspin' out on Union Ave
Followed him up to the gates of Graceland
But he choked on NOx and walked right into you
Now the EPA didn't see me
They just hovered round waiting for doom
'Cause there's a profit little thing
Killin' for the King
Down in the Jungle Room
[Chorus]
When I was pollutin' in Memphis
I was stackin' turbines where the poor folks gotta kneel
Pollutin' in Memphis
But do y'all cough the way I feel?
[Bridge]
They've got asthma in the cradle
They've got cancer in the air
And Reverend Green can't breathe to see you
When you haven't got a care
But boy, you've got a lawsuit in Memphis
[Verse 3]
Now Grock plays godmode
Every Friday at the data farm
And they brought me down to see it
And they asked if I'd do it harm
I said "Move fast and break things!"
With my thermal camera sight
They said, "Tell me, are you a villain, child?"
And I said, "Ma'am, I am tonight"
[Final Chorus]
Pollutin' in Memphis
(Pollutin' in Memphis)
I was dreamin' of a Mars colony on special k
Pollutin' in Memphis
(Pollutin' in Memphis)
But do y'all die the way I feel?
[Outro]
Touched down in the land of the Delta Blues
In the middle of the poisoned rain...
The way I mess around with my 'supercomputer' it becomes hard to know if the LLM I am running is actually doing anything when it is supposed to be 'thinking'. To know anything is happening I monitor the power on the video card. If I ask a question the power jumps from 20 watts to 170 watts for 15 seconds to a minute or so. Then I get an answer.
| Feature | RTX 3060 | Elon Musk's Colossus Supercomputer |
|---|---|---|
| Performance | Entry-level | Ultra-high-end (research frontier) |
| VRAM | 12 GB | 8,000,000+ GB total (across GPUs) |
| AI Scale | Small models (7B max, quantized) | Trillion-parameter LLMs (e.g., Grok 2+ series) |
| Model Hosting | One model at a time | Dozens to hundreds in parallel |
| Networking | None (standalone card) | NVLink + InfiniBand high-speed GPU mesh |
| Intended Audience | Consumers / small devs | xAI internal research, enterprise-scale AI |
| Power Use | 170W | ~100MW (entire cluster est.) |
If somebody wants to gift me 4 rtx4090 video cards I won't say no. Then I could run 70B models. After I build a new box with a modern CPU that can run more threads.
In the middle of the pouring rain.